Artificial intelligence outperforms human classifying of endoscopic images in UC

A deep learning model exceeded human evaluation of classifying endoscopic images in patients with ulcerative colitis (UC). This result shows that artificial intelligence has potential to optimise and standardise the assessment of disease activity in this population. Moreover, the study demonstrated that artificial intelligence surpassed human evaluation with a limited number of images.

Mayo endoscopic subscores (MES) are used to identify disease activity in patients with UC. However, observer variance up to 75% has been reported. Dr Bobby Lo (Hvidovre Hospital, Denmark) and colleagues aimed to develop a deep learning model that is able to distinguish between the 4 MES scores [1].

Initially, 1,484 endoscopic images from 467 UC patients were scored independently by 2 experts. Subsequently, 85% of the images was used as a training set for the machine learning process. The other 15% was used as a test set. The developed deep learning model outperformed human evaluation of the endoscopic images, distinguishing between all MES scores with an accuracy of 84% for the test set, with a sensitivity of 88% and a specificity of 81%. In addition, the deep learning model demonstrated excellent results of distinguishing between inactive to mild (MES 0-1) and moderate to severe (MES 2-3) disease. On the other hand, the weighted kappa value between the 2 expert assessors was 0.66.

Dr Lo mentioned that the model is currently launched for local usage and that the model is under evaluation as a training tool for inexperienced physicians.

1. Lo B, et al. Artificial intelligence surpasses gastrointestinal experts in the classification of endoscopic severity among Ulcerative Colitis. OP07, ECCO 2021 Virtual Congress, 2-3 & 8-10 July.

 

Copyright ©2021 Medicom Medical Publishers



Posted on 2 July, 20218 April, 2024